EP1111364B1 - Damage detection of motor pieces - Google Patents

Description

Field of the invention

The invention lies in the field of permanent monitoring methods of a mechanical part, in particular a large rotating part of an engine, for example a bearing or a gear of an aircraft engine. The method according to the invention relates to monitoring in which the signal from one or more acceleration sensors is processed. The processing of the vibratory signal from the sensor leads to the determination of one or more quantities resulting from the signal or derived from the signal by the processing. These quantities are compared with thresholds to detect that the monitored mechanical part has been damaged.

State of the art

It is known to have acceleration sensors at locations where these sensors can detect vibrations from mechanical parts to be monitored or the machine in general, and then process the signals from these sensors to detect a significant anomaly of the signal transmitted by the sensor with respect to the signal received in the absence of a fault.

Ideally, the detection of such anomalies should allow a life expectancy remaining part of the part under surveillance before serious damage leading to a serious rupture or malfunction.

In practice, there is usually not enough experience on a large enough number of pieces to obtain statistically significant data to arrive at such a forecast.

In practice, the result of the treatment is such that damage can be detected early enough so that at the time of detection the organ still functions satisfactorily and there is reasonable assurance that will continue to operate satisfactorily, preferably until a next periodic visit of this organ and at least until a next stopover when the body is mounted on a flying machine.

Requests for EP 0 889 313 316 A2 filed simultaneously on July 3, 1998, provide a good example of such methods.

In these applications, fifteen acceleration sensors and two azimuth sensors for sensing rotational speeds are arranged in various places near members that together make up a unit for transmitting movements to rotary members of a helicopter, particular to the rotor shaft and the rear stabilizer propeller.

• une étape d'acquisition du signal d'un capteur d'accélération ; il s'agit là d'une étape de numérisation du signal du capteur réalisée par exemple au moyen d'un échantillonneur bloqueur et d'un convertisseur analogique-numérique ;
• une étape de transformation du signal temporel ainsi capté, en un signal dans le domaine fréquentiel, ceci pour chacune des trois demandes '313, '314 et '315 ; on obtient ainsi une séquence initiale de valeurs déterminant chacune une fréquence de vibration et l'amplitude associée à cette fréquence ; le traitement décrit ensuite dans la demande '313 prévoit une sélection et un traitement des échantillons fréquentiels pour obtenir une séquence finale d'échantillons.

All of the signal processing methods described in these four applications include:

• a step of acquiring the signal of an acceleration sensor; this is a step of digitizing the sensor signal carried out by example by means of a sample-and-hold device and an analog-to-digital converter;
• a step of transforming the time signal thus captured, into a signal in the frequency domain, for each of the three requests 313, 314 and 315; an initial sequence of values is thus obtained, each determining a vibration frequency and the amplitude associated with this frequency; the treatment described in the '313 application then provides selection and processing of the frequency samples to obtain a final sequence of samples.

After a return to the time domain, starting from the final sequence of samples, a moment of order 6 of this temporal signal is computed which is compared with a predetermined threshold to possibly actuate an alarm. The method described in the '313 application is intended to predict the arrival of an anomaly located at an outer shaft.
The processing described in the '314 application then provides for the selection of a determined sample of frequency, the calculation of the amplitude of this sample and its comparison with a reference value, the result being compared with a threshold.

This relatively frustrating method is intended to detect anomalies developing rapidly in flight.

The treatment described next in the '315 application provides for the selection of two groups of frequency samples, the calculation of the energy associated with these two groups, an energy gap calculation between the two groups and the comparison of this difference to a threshold.

This method is intended to detect a fault on a shaft having two gears.

The processing described in the '316 application provides meanwhile after the acquisition phase, a Hilbert transformation of the signal obtained, the definition of a complex number having the signal for real part and the Hilbert transform for imaginary part, the calculating the phase of this complex number and its derivative with respect to time and finally comparing this derivative to a threshold value.

In each of these four requests, one looks for ratios or variations intervening on what is called "a frequency of meshing" or harmonics of this frequency. It is assumed that this is the number of revolutions per second of the gear under surveillance.

The reasons for the size of these requests, rather than others, are not explained, which makes it unclear to what extent the teaching of these requests can be used in a different context. from that which is described.

The teaching that can be drawn from these examples is that the analysis of vibration signals of sensors placed near rotating parts can give indications on the mechanical state of these parts. In particular, a crack initiation or crack development can be identified by such analysis. However, for each particular case, it is necessary to determine which are the frequencies to be analyzed and among all the processing possibilities what are the most significant quantities to monitor to have significant information on the monitored mechanical part.

Brief description of the invention

The invention is intended to detect early damage particularly occurring to a rotating part of a motor for example a ball bearing or rollers, or a gear. However, it is possible to extend the scope of the invention to a non-rotating part, for example a fixed part, for example a cowling or mobile according to another movement, for example a connecting rod or a valve and its rod. According to the invention, as in the prior art, a digital capture of the signal delivered by at least one acceleration sensor dedicated to vibratory monitoring of the motor is carried out. This time signal is then, as in the prior art, transformed in the frequency domain. It has been seen that in the prior art a filtering is then carried out to select characteristic frequencies of the desired defect which have been called in the cited patents "meshing frequencies". Treatments are then performed on these characteristic frequencies to obtain quantities that can be compared to thresholds to draw conclusions relating to a possible damage to the mechanical part followed.

The inventors of the present invention have noticed that the frequency signal obtained from the transformation of the time signal coming from the sensor is very noisy. They have also noticed that the signal corresponding to what has been called the gearing frequency is very weak in the sensor, so that it is difficult to isolate the possible presence of this gearing frequency of the gear. ambient noise necessarily present on an engine.

The inventors have also noticed that this "meshing frequency" can be present even in the absence of defects, simply because the teeth mesh with each other. What must be detected is therefore a significant modification of the spectrum associated with this meshing frequency, for example a modification of the number of harmonic lines of this frequency and / or the amplitude of vibration at this meshing frequency or to his harmonics.

They also hypothesized that this meshing frequency could be present as the frequency of modification of one or more fundamental frequencies of the engine. By fundamental frequencies of the motor, denotes the rotation frequency - number of revolutions per second - of the engine, or for the engines having a low pressure compressor and a high pressure compressor, the rotation frequencies of each of these bodies. In general, a fundamental frequency of a motor is a rotation frequency of a large part, from the moment of inertia point of view, of the motor. A frequency resulting from damage will be a frequency at which this damage produces a shock. Ideally, the frequency of engraining or rolling would result of a succession at this pulse frequency of DIRAC.

In practice, these pulses have a less pure form than a DIRAC pulse. The spectrum of the fault signal is therefore composed of lines at the frequency corresponding to the nature of the fault and to harmonic frequencies. The envelope of this spectrum of. rays is determined by the shape of the pulses. It will be seen that in the case of the search for damage to a bearing ring the specific damage frequency of the presence of the defect represents, per unit of time, the number of times that this defect is struck by a ball or a roll of the bearing. In the case of a gear of which for example a tooth of a first wheel would be cracked or damaged, the frequency resulting from the damage would be the frequency with which this tooth comes into contact with the teeth of a second wheel meshing with the first. We see by these examples that, knowing the structure of the motor, its speed of rotation, so its instantaneous fundamental frequencies, it is possible to know a frequency resulting from the damage, emitted by the piece which one monitors.

The problem is therefore to detect the presence or the modification of characteristics of this frequency resulting from the damage in a sufficiently stable and constant manner to isolate it from the noise, and to conclude that there is a defect in the part being tracked, this without generating false alarms. It is also important to identify this frequency resulting from the damage while the defect causing the the appearance of this frequency is still at an early stage of its development. For example for a bearing, an early stage consists of a simple peeling of one of the rolling surfaces. However, at this early stage, the amplitude of the frequency or the modification of the characteristics of its associated spectrum resulting from the defect is small so that it is difficult to distinguish them from the noise.

According to the invention, as in the prior art, a sequence of the signal of at least one sensor of a vibratory acceleration signal is sampled, with a sampling frequency sufficient for Shannon's condition to be fulfilled for the first time. higher frequencies to record. Next, according to a characteristic of the present invention, it is verified that the engine speed has remained stable throughout the duration of the sequence, so as to reject the sequences for which the relative variation of the rotational speed is above a predetermined threshold. Then, as in the prior art, converting the time signal obtained into a frequency signal, for example by Fourier transformation and remove the fundamental frequencies of the engine.

Then, according to an important characteristic of the invention, a correlation will be made between the frequency signal obtained and the same signal shifted by a frequency V.

The correlation of two functions is the measure of the similarity between them: it is essentially a method of comparison.

In digital, the signals are represented by series of numbers that are successive samples of a continuous waveform.

avec i = indice de l'échantillon, et k = indice de décalage.The correlation function of f (k) and g (k), the two numerical sequences obtained by sampling two continuous functions f (t) and g (t), is expressed mathematically by the relation: $$S\left(k\right)={\int }_{i=-\infty }\mathrm{f}\left(\mathrm{i}\right)\mathrm{.}\mathrm{boy\; Wut}\left(\mathrm{i}\mathrm{+}\mathrm{k}\right)$$

with i = index of the sample, and k = index of offset.

When f and g are different functions, we speak of intercorrelation.

When f and g are identical functions, we speak of autocorrelation.

In the case of a frequency signal expressed in the amplitude / frequency complex plane, spectrum and spectral coherence are referred to.

The spectral coherence calculation between the frequency domain transform of the picked vibratory signal and this same offset transform of a frequency V makes it possible to identify the similarities between the two transforms. Under these conditions, the noise frequencies, unstable by nature will be eliminated to leave only stable frequencies, that is to say those that result from a stable origin.

In this case they are the fundamental frequencies, their harmonics, the linear combinations of these frequencies and any stable modulation frequencies.

These stable modulation frequencies may include, even in the absence of a fault, the meshing frequency or in the case of a rolling bearing, a rolling frequency resulting from the number of balls or rollers and the rotation frequency of the set formed by these balls or rollers.

If, for example, F _N1 and F _N2 are the high mass rotation frequencies of the motor, for example the rotors of a low pressure body and a high pressure body respectively, the signal delivered by the sensor will comprise the fundamental frequencies F _N1. and F _N2 . It could also include the harmonic frequencies of these frequencies for example _N1 2F, 3F _{N1, N2} 2F, 3F _2. If, furthermore, the transmission chain is such that a signal comprising the frequency F _N1 passes through a vibrating member, for example at the frequency 2F _N2 , there will be frequencies resulting from a modulation of one frequency by another, for example 2F _N2 + F _N1 , 2F _N2 - F _N1 . For these reasons, the spectrum of an engine will present for each particular case, that is to say for each type of engine and each sensor location and also for ranges of engine speed characteristic spectrum lines.

For a type of motor, a sliding coherence calculation, that is to say by varying the shift V step by step from frequency values resulting from combinations of fundamental values obtained on small multiples of the fundamental frequencies for example the included multiples, limits included, between 0 and 4 and around these combinations, will make it possible to locate for the engine regimes studied, for example for the regimes the longest used during each flight, the lines present normally, that is to say in the absence of defects. It will then be necessary to perform the same search for lines with an engine equipped with a part having at an early stage the defect that will then be sought to detect.

For this search, the offset V will be set to frequency values resulting from the modulation of the frequencies present in normal operation by the frequency resulting from the damage or by harmonics of this frequency.

For the method according to the invention to be of interest, the spectrum resulting from the defect must be different from the spectrum in normal operation, by the presence of a frequency which is not present with a sufficient amplitude to be detected. even by means of the spectral coherence calculation. Indeed, a possible modification of the amplitude of this frequency due to the presence of the defect does not necessarily cause a significant variation in the value of the coherence peak.

On the other hand, such a modification of the amplitude of a frequency due to the defect can be detected by known methods comprising bandpass filtering centered on the desired frequency and a calculation of the energy or amplitude of the resulting spectrum. filtering.

• une phase préliminaire d'identification de raies spectrales fréquentielles présentes au cours du fonctionnement du moteur en l'absence de défaut, puis en présence du défaut de façon à identifier des raies spectrales spécifiques du défaut, ceci pour au moins un régime de fonctionnement du moteur,
• une phase de détection au cours de laquelle de façon. itérative au cours du fonctionnement du moteur :

• on acquiert au cours d'une séquence d'acquisition une suite d'échantillons numériques représentative d'un signal vibratoire d'accélération,
• on vérifie qu'au cours de ladite séquence d'acquisition le régime du moteur est resté stable et on applique en temps réel ou en temps différé auxdites séquences stables acquises les traitements ci-après,
• on transforme le signal acquis par échantillonnage temporel en un signal fréquentiel, en éliminant les fréquences fondamentales du moteur et leurs harmoniques,
• on effectue au moins un calcul de cohérence normée entre ledit signal fréquentiel et le même signal fréquentiel décalé d'une valeur de fréquence correspondant à la valeur d'une des fréquences spécifiques du défaut, détectée au cours de la phase préliminaire,
• on compare la valeur d'un pic de cohérence obtenu par le calcul de cohérence à un premier seuil de cohérence et on mémorise une valeur de détection 1 si ce pic est supérieur à ce premier seuil et une valeur 0 dans le cas contraire,
• on fait la somme de p valeurs de détection mémorisées que l'on divise par p pour obtenir un ratio de détection,
• on décide qu'un défaut est présent si le ratio de détection est supérieur ou égal à un second seuil prédéterminé.

In summary, the invention relates to a method for early detection of the appearance of a defect in a part of an engine, the method comprising:

• a preliminary phase of identification of frequency spectral lines present during the operation of the motor in the absence of defect, then in the presence of the defect so as to identify specific spectral lines of the defect, this for at least one operating speed of the engine ,
• a detection phase during which way. iterative during engine operation:

• a sequence of digital samples representative of a vibratory acceleration signal is acquired during an acquisition sequence,
• it is verified that, during said acquisition sequence, the engine speed has remained stable and the said treatments are applied in real time or in deferred time to the said stable sequences acquired,
• the signal acquired by temporal sampling is transformed into a frequency signal, eliminating the fundamental frequencies of the motor and their harmonics,
• at least one normalized coherence calculation is carried out between said frequency signal and the same frequency signal shifted by a frequency value corresponding to the value of one of the specific frequencies of the fault, detected during the preliminary phase,
• comparing the value of a coherence peak obtained by the coherence calculation with a first coherence threshold and storing a value of detection 1 if this peak is greater than this first threshold and a value 0 in the opposite case,
• sum of p stored detection values which are divided by p to obtain a detection ratio,
• it is decided that a fault is present if the detection ratio is greater than or equal to a second predetermined threshold.

Preferably, the p detection values used to establish the detection ratio are values corresponding to P selected from the most recent sequences recorded in engine speed ranges considered to be of interest. It will not necessarily be the most recent in time absolutely, but more recent in these ranges of engine speed that we consider particularly interesting.

The number P is an integer greater than or equal to 1.

Brief description of the drawings

• la figure 1 représente schématiquement une coupe longitudinale d'un roulement de palier,
• la figure 2a représente une coupe schématique partielle d'un roulement par un plan perpendiculaire à l'axe du roulement, la figure 2b par un plan axial du roulement.
• la figure 3 est un organigramme schématisant un mode préféré de réalisation de l'invention

Other features of the method according to the invention and variants will be explained in the description of embodiments which will be described below with reference to the accompanying drawings in which:

• the figure 1 schematically represents a longitudinal section of a bearing bearing,
• the figure 2a represents a partial schematic section of a bearing by a plane perpendicular to the axis of the bearing, the figure 2b by an axial plane of the bearing.
• the figure 3 is a flowchart schematizing a preferred embodiment of the invention

Description of exemplary embodiments

The method according to the invention has been used for monitoring the state of a rotor bearing of an engine. This bearing is shown very schematically in axial section on the figure 1 .

The bearing comprises a roller bearing 1. The rollers roll between two rings, an inner ring 3 and an outer ring 4.

The inner ring 3 centers a shaft 5 of a rotor of a low-pressure body of the engine. This body is not otherwise represented because it is not useful for understanding the invention. The outer ring 4 centers a shaft 6 of a high pressure body of the same motor.

Partial transverse and axial sections of this bearing are shown Figures 2a and 2b respectively. Trees 5 and 6 ( figure 1 ) rotate at different speeds of rotation, W _e for the shaft 6 centered by the outer ring 4 and W _i for the shaft 5 centered by the inner ring 3. It can be considered that the rotation speeds W _e and W _i are independent of each other. They are however known at all times. The outer ring connected to the rotor of the high pressure body rotates faster than the inner ring and the two rings rotate in the same direction.

Between the two rings are Z rolling bodies 2, for example rollers, held by a cage 7. With the slip close, low steady state, the cage 7 rotates at a speed W _c . A rolling factor γ is defined as being the ratio between the diameter d of a roll with the diameter dm of the median cylinder of the two rings 3 and 4.

The defect that is to be detected is a defect in the outer ring 4 of the bearing, for example chipping. This defect is represented by a star on the figure 2a .

The rotation speed W _c of the cage is given by the formula: $${\mathrm{W}}_{\mathrm{vs}}\mathrm{=}\frac{\mathrm{1}\mathrm{-}\mathrm{<figure-callout\; id="2"\; label="\gamma "\; filenames="imgf0001.png,imgf0002.png"\; state="\{\{state\}\}">\gamma </figure-callout>}}{\mathrm{2}}{\mathrm{W}}_{\mathrm{i}}+\frac{\mathrm{1}+\mathrm{<figure-callout\; id="2"\; label="\gamma "\; filenames="imgf0001.png,imgf0002.png"\; state="\{\{state\}\}">\gamma </figure-callout>}}{\mathrm{2}}{\mathrm{W}}_{\mathrm{e}}$$

The frequency f _D resulting from the damage will be equal to the number of times a roll hits the defect.

If one neglects a sliding factor of the rolling bodies, this frequency fD is proportional to the difference N2 -N ₁ between the number of revolutions per second of the outer ring and the number of revolutions per second of the inner ring.

The hypothesis emitted by the inventors is that this frequency f _d will be transmitted to the sensor of a vibratory acceleration signal through organs that are themselves vibrating, in particular at fundamental frequencies.

A coherence calculation is performed between the Fourier transform of the picked vibratory signal and the Fourier transform of the same signal shifted by a cyclic frequency V.

The asymmetric spectral coherence of a signal x (t) for the cyclic frequency V is given by the formula: $${\mathrm{VS}}_{\mathrm{x}}^{\mathrm{v}}\left(\mathrm{f}\right)\mathrm{=}\frac{\mathrm{E}\left[\mathrm{x}\left(\mathrm{f}\right)\mathrm{.}\mathrm{x}\mathrm{*}\left(\mathrm{f}\mathrm{-}\mathrm{V}\right)\right]}{{\mathrm{(}\mathrm{E}[{\left|\mathrm{<figure-callout\; id="2"\; label="x\; f"\; filenames="imgf0001.png,imgf0002.png"\; state="\{\{state\}\}">x\; </figure-callout>}\left(\mathrm{f}\right)\left(\mathrm{}\right)\right|}^{\mathrm{2}}\mathrm{.}\mathrm{E}[{\left|\mathrm{x}\left(\mathrm{f}\mathrm{-}\mathrm{V}\right)\right|}^{\mathrm{2}})}^{1/2}}$$

• ${\mathrm{C}}_{\mathrm{x}}^{\mathrm{v}}\left(\mathrm{f}\right)$
  
  est la valeur de cohérence spectrale entre le signal vibratoire capté par le capteur et le même signal décalé de la fréquence cyclique V,
• X(f) représente la transformée de Fourrier du signal vibratoire X(t) X^*(f-v) représente la fonction conjuguée de la transformée du signal vibratoire x(t) décalée de la fréquence cyclique V,
• E [.....] désigne l'espérance mathématique,
• E [|X(f)|²] et E [|X(f - v)|²] représentent respectivement les densités spectrales de puissance des signaux X(t) et de la transformée de Fourrier inverse de X(f-v),
• E [X(f) .X * (f - v) représente la densité spectrale d'interaction des signaux x(t) et de la transformée de Fourrier inverse de X(f-v).

In this formula

• ${\mathrm{VS}}_{\mathrm{x}}^{\mathrm{v}}\left(\mathrm{f}\right)$
  
  is the value of spectral coherence between the vibratory signal picked up by the sensor and the same signal shifted from the cyclic frequency V,
• X (f) represents the Fourier transform of the vibratory signal X (t) X ^* (fv) represents the conjugate function of the vibration signal transform x (t) shifted from the cyclic frequency V,
• E [.....] denotes the mathematical expectation,
• E [| X (f) | ² ] and E [| X (f - v) | ² ] respectively represent the power spectral densities of the signals X (t) and the inverse Fourier transform of X (fv),
• E [X (f) .X * (f - v) represents the interaction spectral density of the signals x (t) and the inverse Fourier transform of X (fv).

Further information can be found in HURD "An investigation of Periodically correlated processes" Ph. D dissertation in engineering. DUKE UNIVERSITY 1970 and in GARDNER "Statistical Spectral Analysis, A Non Probabilistic Theory" Prentice Hall 1988.

The estimation method used is that of the averaged periodogram that uses the fast Fourier transform. Max and LACOUME "Methods and signal processing techniques and applications to physical, Volume 1: General principles and conventional methods" ^5th edition, 1996 MASSON.

The spectral coherence is therefore normed between 0 and 1. Given the fact that we work on a signal of finite duration and more noisy, we can not have values equal to 0 or 1. A strong spectral coherence, by example greater than 0.8 for the frequencies f and (fv) indicates that these frequencies are derived from the same physical phenomenon, for example by modulation of a source signal for example, at the frequency F _N1 or F _N2 by a periodic signal at the fault frequency F _D = (F _N2 - F _N1 ).

Depending on the position of the sensor with respect to the fault, there may be modulations due to unbalance of the shafts 6 or 5 causing modulation by a frequency F _N2 and / or F _N1 respectively. In the particular case studied, it was noted that the modulation frequencies of the fault frequency were functions of the difference (F _N2 - F _N1 ) of rotation frequency of the two shafts.

• une séquence de prise d'échantillons ; cette séquence peut avoir une durée comprise par exemple entre 30 secondes et une minute ;
• la vérification que pendant la durée de la séquence le régime est resté stable, et le rejet des séquences ne répondant pas à cette condition.

Under these conditions, the permanent search for the appearance of the fault based on the search for the fault frequency includes iteratively as explained above:

• a sampling sequence; this sequence can have a duration of for example between 30 seconds and a minute;
• the verification that during the duration of the sequence the regime remained stable, and the rejection of the sequences not satisfying this condition.

It should be noted that, alternatively, the sampling sequence can be started only if the engine is running at a speed within a range that has been determined during the preliminary phase that it was of interest. particular.

If the sequence satisfies the conditions of stability required, the treatment according to the invention is begun.

According to the embodiment described here, the sequence is divided into n parts, so as to obtain, for example, subsequences of one second to a few seconds each. In the preferred embodiment, the subsequences overlap each other partially, for example by 50%.

A discrete Fourier transform of each subsequence is then performed.

The coherence calculation is carried out on each of the subsequences taking, as the offset value V, the different modulation frequencies resulting from the transmission model described above.

Each coherence calculation makes it possible to obtain n coherence samples. The average consistency is equal to the average of the consistency samples.

If for at least one of the offsets, the normed average coherence is greater than a threshold of 0.8, there is coherence and it is memorized. The treatment is then continued as described above.

When for the p preferably last consecutive sequences and for the same offset, the number of positive coherence sequences is greater than a predetermined threshold, or which amounts to the same if the ratio of the number of positive coherences to the number p of sequences is greater than a threshold, it is decided that there is a defect .

Alternatively or additionally, for each sequence within a succession of sequences, it is possible to carry out a sliding coherence calculation, and to memorize the frequency values V for which a consistency peak greater than the predetermined threshold is obtained by example 0.8. This is a signature of the regime in which the sequence was performed. This signature is compared to an earlier stored signature.

Sequences for which there is a significant difference between the current signature and the stored signature are counted. It is decided that there is an anomaly if for a number Q of consecutive sequences, there is a number r of sequences deviating from the stored signature greater than a predetermined threshold (r ≤ Q).

The experiments conducted by the Applicant on different engines of the same type have found that in some cases the consistency calculation does not always lead to a detection of the defect while it is detectable by other methods. Other times the defect is detected by the consistency method while it is not detected by other methods. This is why, according to a preferred method of defect detection allowing more emission and differentiation of alert levels is carried out additional treatments. Some of these treatments can be performed in parallel with the treatment described above.

For these processing after transformation of the digital time signal into a frequency signal, for example by fast Fourier transform and filtering of the fundamental frequencies of the motor and their harmonics, it is examined whether the highest amplitudes of the frequency signal are greater than a first low threshold. amplitude and if so, whether they are also greater than a second high threshold of amplitudes. If the highest amplitudes are greater than the low threshold, it is examined whether on several data sequences, for example "a" data sequences, the degree of differential harmonic is the same for at least b sequences, a and b are integers greater than or equal to 1, b ≤ a. The degree or ratio of differential harmonic is the value of the ratio between a frequency value of amplitude greater than one of the thresholds and the value of the fault frequency.

In the case of rolling, this frequency is (F _N2 -F _N1 ) as explained above.

If the values of these ratios remain constant at a predetermined error percentage, it will be concluded that the defect is present.

A level 1 message will be issued if the harmonic ratio remains constant for at least b sequences among sequences and the amplitude threshold is greater than the second high amplitude threshold. This will mean that the fault is present and sufficiently well marked to cause this high level of amplitude.

A level 2 message will be issued if the harmonic ratio remains constant for at least b sequences among sequences and the amplitude level is greater than the second low threshold while being lower than the second high amplitude threshold.

1. 1) si comme indiqué plus haut le test de cohérence spectrale effectué avec V = F_N2 - F_N1 conduit à un ratio de détection supérieur ou égal au second seuil de ratio de détection prédéterminé,
2. 2) si comme indiqué plus haut, le test de cohérence spectrale effectué avec V = N₁ ou N₂ conduit à un ratio de détection supérieur ou égal au second seuil de ratio prédéterminé et que de plus les pics de cohérence sont espacées d'une valeur (f_N2-f_N1) pour au moins b séquences parmi a,
3. 3) si comme indiqué en 2) ci-dessus, le test de cohérence spectrale effectué avec V = N₁ ou N₂ conduit à un ratio de détection supérieur ou égal au second seuil de ratio prédéterminé et que de plus le ratio d'harmonique différentielle entre l'une des fréquences pour laquelle un pic de cohérence a été détecté et la fréquence de défaut est stable pour plusieurs séquences traitées, c'est-à-dire reste constant pour b séquences parmi a séquences.

A level 3 message will be issued if the amplitude level remains below the first low level threshold and the spectral consistency tests give one of the following results:

1. 1) if, as indicated above, the spectral coherence test carried out with V = F _N2 -F _N1 leads to a detection ratio greater than or equal to the second predetermined detection ratio threshold,
2. 2) if, as indicated above, the spectral coherence test carried out with V = N ₁ or N ₂ leads to a detection ratio greater than or equal to the second predetermined ratio threshold and that moreover the coherence peaks are spaced apart by value (f _N2 -f _N1 ) for at least b sequences out of a,
3. 3) if as indicated in 2) above, the spectral coherence test carried out with V = N ₁ or N ₂ leads to a detection ratio greater than or equal to the second predetermined ratio threshold and that also the harmonic ratio differential between one of the frequencies for which a coherence peak has been detected and the fault frequency is stable for several sequences treated, that is to say remains constant for b sequences among sequences.

The treatment and alerting mode described above is summarized in figure 3 .

In a step 10, one proceeds to the acquisition and the backup for example by storing data from a sensor of a vibratory acceleration signal. As already explained above, the sequences for which the engine speed is unstable during the recording of the frequency are not retained for saving.

In one step 11 Fast Fourier Transformation (FFT) and filtering are performed to eliminate the frequencies resulting from the rotation of high kinetic momentum rotor assemblies such as high and low pressure bodies.

In a step 12 spectral coherence calculations are carried out with offsets V corresponding to the fault frequency, or to each of the frequencies resulting from the rotation of the high and low pressure bodies. In a step 13, a level 3 message is optionally emitted if one of the results 1, 2 or 3 described above is obtained.

In parallel with step 12, after step 11 of transforming into a frequency signal and filtering, the amplitude level of the frequencies of the highest amplitudes of the frequency signal at a first low threshold is compared in a step 14.

If in a manner deemed stable, because appearing on a number of recorded sequences equal to 3 (b = a = 3) frequency amplitudes are greater than the first low threshold, comparing in a step 15 these amplitudes to a second high threshold.

If in step 14 it is found that the largest amplitudes of the frequency signal are lower than at the first low threshold, parallel processing stops. In such a case, the treatment comprises only steps 10, 11, 12 and possibly 13.

If, on the other hand, it is found that the largest amplitudes of the frequency signal are stably higher than the first low threshold, a step 17 will be considered if the degree of differential harmonic of at least one of the frequencies above the low threshold. is stable. If so, a level 2 message will be transmitted in step 18. If, moreover, the largest amplitudes of the signal are greater than the second high threshold, and it is noted in step 17 that the degree of differential harmonic is stable then one will issue in a step 19 a level 1 message.

The degree or ratio of differential harmonic is obtained by dividing the frequency of the spectrum having a high amplitude by the fault frequency (F _N2 - F _N1 ).

The threshold value of the coherence peak, above which it is considered that there is coherence, is fixed at 0.8.

In the particular example shown in figure 3 In summary, according to the preferred embodiment of the invention, a level 3 message is issued only when the detection of the fault results only from the coherence calculation.

This message is issued if with V = (f _N2 -f _N1 ) the ratio between the number of times that we have a coherence peak greater than the first threshold, for example 0.8, and the number p of the sequences examined is greater than the second predetermined threshold.

If the detection is obtained by making the coherence computation with F _N1 or f _N2 one looks further if among sequences examined during this computation there are at least b among them for which the ratio of differential harmonic is constant .

The numbers p, a and b and the second predetermined threshold are independent of one another. However by construction, the second predetermined threshold is less than 1, and b ≤ a.

Level 2 and 3 messages are transmitted if stably, i.e., for b sequences at least taken from among the amplitudes are greater than the first low threshold or the first high threshold. Obviously, if an amplitude is greater than the high threshold, it is necessarily greater than the low threshold. It may therefore happen that a level 2 message and a level 1 message must be transmitted simultaneously. In this case, only the level 1 message will be transmitted.

In step 17, the frequencies to be taken into account for performing the calculation of the differential harmonic ratio can be determined by a spectral coherence calculation with V = (f _N2 - f _N1 ). The difference with step 12 cases 1 or 2 is that in step 17 we look at the constancy of the differential harmonic ratio.

Claims (9)

1. Process for the early detection of the appearance of a fault in a component of an engine, the process comprising:

   - a preliminary phase of investigations to identify frequency spectral lines present in the course of the operation of an engine of like design in the absence of a fault, and in the presence of a fault so as to identify specific spectral lines of the fault, doing so for at least one speed regime of operation of the engine,

   - a phase of detection in the course of which in an iterative manner in the course of the engine operation:

   - a string of digital samples which is representative of an acceleration vibrational signal is acquired in the course of an acquisition sequence,

   - one checks that in the course of said acquisition sequence the speed regime of the engine has remained stable and the processing operations hereinbelow are applied in real time in the course of the operation of the motor or in non-real time to said stable sequences acquired,

   - one transforms the signal acquired by temporal sampling into a frequency signal, while eliminating the fundamental frequencies of the engine and their harmonics,

   - at least one normed coherence calculation is performed between said frequency signal and the same frequency signal shifted by a frequency value corresponding to the value of one of the specific frequencies of the fault, and which is detected in the course of the preliminary phase,

   - the value of a coherence peak obtained via the coherence calculation is compared with a first threshold and a 1 detection value is stored if this peak is greater than this first threshold and a 0 value in the contrary case,

   - p stored detection values are summed and the sum is divided by p to obtain a detection ratio,

   - a fault is deemed to be present if the detection ratio is greater than or equal to a second predetermined threshold.
2. Process according to Claim 1, in which spans of interest are defined and one checks prior to the execution of a samples acquisition sequence that the speed regime of the engine is in one of the spans of interest.
3. Process according to either of Claims 1 and 2, characterized in that the p detection values used to establish the value of the detection ratio result from the processing of acquired sequences which are the most recent in each of the engine speed regime spans in the course of which these sequences were acquired.
4. Process according to Claim 1, characterized in that, in addition to the detection of a fault, a detection message having a level is issued, the level being variable as a function of detection-related conditions.
5. Process according to Claim 4, characterized in that a message of level 3 is issued if the spectral coherence peak is obtained for a frequency shift V corresponding to an unmodulated frequency (F_N2 - F_N1) specific to the frequency of the fault, F_N1 and F_N2 being the frequencies of rotation of significant masses of the engine.
6. Process according to Claim 4, characterized in that a message of level 3 is issued if spectral coherence peaks are obtained via a frequency shift V corresponding to one of the fundamental frequencies of the engine and, characterized in that moreover, coherent peaks of level greater than the threshold are spaced apart by a frequency value corresponding to an unmodulated frequency (F_N2 - F_N1) specific to the fault frequency, F_N1 and F_N2 being the frequencies of rotation of significant masses of the engine.
7. Process according to Claim 5, characterized in that a message of level 3 is issued if one or more coherence peaks are obtained with a frequency shift V corresponding to one of the fundamental frequencies of the engine and, characterized in that moreover, the ratio between the frequency for which at least one of the coherence peaks is greater than the second coherence threshold and one of the frequencies corresponding to an unmodulated frequency (F_N2 - F_N1) specific to the presence of the fault remains the same for a minimum number b of sequences out of a predetermined number a of sequences.
8. Process according to one of Claims 4 to 7, characterized in that, in parallel with the coherence processing, amplitudes of frequencies making up the frequency signal obtained after transforming the acceleration vibrational temporal signal into a frequency signal are compared with a first low threshold, and characterized in that if for one of the frequencies whose amplitude is greater than this first low threshold the ratio between the frequency corresponding to this amplitude greater than the first low threshold of amplitude and one of the frequencies corresponding to an unmodulated frequency (F_N2 - F_N1) specific to the presence of the fault remains the same for a predetermined number of sequences, a message of level 2 is issued.
9. Process according to one of Claims 4 to 7, characterized in that, in parallel with the coherence processing, amplitudes of frequencies making up the frequency signal obtained after transforming the acceleration vibrational temporal signal into a frequency signal are compared with a second high threshold, and characterized in that if for one of the frequencies whose amplitude is greater than this second high threshold the ratio between the frequency corresponding to this amplitude greater than the second high threshold of amplitude and one of the frequencies corresponding to an unmodulated frequency (F_N2 - F_N1) specific to the presence of the fault remains the same for a predetermined number of sequences, a message of level 1 is issued.

Images